Structural Determinants of Vertebral Fracture Risk

Authors

  • L Joseph Melton III MD,

    Corresponding author
    1. Division of Epidemiology, Department of Health Sciences Research, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    2. Division of Endocrinology, Metabolism, and Nutrition, Department of Internal Medicine, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    • Division of Epidemiology, Department of Health Sciences Research, Mayo Clinic, 200 First Street Southwest, Rochester, MN 55905, USA
    Search for more papers by this author
  • B Lawrence Riggs,

    1. Division of Endocrinology, Metabolism, and Nutrition, Department of Internal Medicine, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    Search for more papers by this author
  • Tony M Keaveny,

    1. University of California, Berkeley, California, USA
    2. O.N. Diagnostics, Berkeley, California, USA
    Search for more papers by this author
    • Dr Keaveny has a financial interest in O.N. Diagnostics, and both he and the company may benefit from the results of this research. Mr Hoffmann has equity interests in and is an employee of O.N. Diagnostics. The other authors state that they have no conflicts of interest.

  • Sara J Achenbach,

    1. Division of Biostatistics, Department of Health Sciences Research, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    Search for more papers by this author
  • Paul F Hoffmann,

    1. O.N. Diagnostics, Berkeley, California, USA
    Search for more papers by this author
  • Jon J Camp,

    1. Biomedical Imaging Resource, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    Search for more papers by this author
  • Peggy A Rouleau,

    1. Division of Computed Tomography, Department of Radiology, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    Search for more papers by this author
  • Mary L Bouxsein,

    1. Orthopedic Biomechanics Laboratory, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA
    Search for more papers by this author
  • Shreyasee Amin,

    1. Division of Rheumatology, Department of Internal Medicine, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    Search for more papers by this author
  • Elizabeth J Atkinson,

    1. Division of Biostatistics, Department of Health Sciences Research, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    Search for more papers by this author
  • Richard A Robb,

    1. Biomedical Imaging Resource, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    Search for more papers by this author
  • Sundeep Khosla

    1. Division of Endocrinology, Metabolism, and Nutrition, Department of Internal Medicine, Mayo Clinic College of Medicine, Rochester, Minnesota, USA
    Search for more papers by this author

Abstract

Vertebral fractures are more strongly associated with specific bone density, structure, and strength parameters than with areal BMD, but all of these variables are correlated.

Introduction: It is unclear whether the association of areal BMD (aBMD) with vertebral fracture risk depends on bone density per se, bone macro- or microstructure, overall bone strength, or spine load/bone strength ratios.

Materials and Methods: From an age-stratified sample of Rochester, MN, women, we identified 40 with a clinically diagnosed vertebral fracture (confirmed semiquantitatively) caused by moderate trauma (cases; mean age, 78.6 ± 9.0 yr) and compared them with 40 controls with no osteoporotic fracture (mean age, 70.9 ± 6.8 yr). Lumbar spine volumetric BMD (vBMD) and geometry were assessed by central QCT, whereas microstructure was evaluated by high-resolution pQCT at the ultradistal radius. Vertebral failure load (∼strength) was estimated from voxel-based finite element models, and the factor-of-risk (ϕ) was determined as the ratio of applied spine loads to failure load.

Results: Spine loading (axial compressive force on L3) was similar in vertebral fracture cases and controls (e.g., for 90° forward flexion, 2639 versus 2706 N; age-adjusted p = 0.173). However, fracture cases had inferior values for most bone density and structure variables. Bone strength measures were also reduced, and the factor-of-risk was 35–37% greater (worse) among women with a vertebral fracture. By age-adjusted logistic regression, relative risks for the strongest fracture predictor in each of the five main variable categories were bone density (total lumbar spine vBMD: OR per SD change, 2.2; 95% CI, 1.1–4.3), bone geometry (vertebral apparent cortical thickness: OR, 2.1; 95% CI, 1.1–4.1), bone microstructure (none significant); bone strength (“cortical” [outer 2 mm] compressive strength: OR, 2.5; 95% CI, 1.3–4.8), and factor-of-risk (ϕ for 90° forward flexion/overall vertebral compressive strength: OR, 3.2; 95% CI, 1.4–7.5). These variables were correlated with spine aBMD (partial r, −0.32 to 0.75), but each was a stronger predictor of fracture in the logistic regression analyses.

Conclusions: The association of aBMD with vertebral fracture risk is explained by its correlation with more specific bone density, structure, and strength parameters. These may allow deeper insights into fracture pathogenesis.

INTRODUCTION

Areal BMD (aBMD) assessed by DXA at the lumbar spine is a mainstay of clinical practice because the measurement can predict subsequent vertebral fractures with a relative risk of about 2.3 per SD change, which is better than the ability of blood pressure measurements to predict stroke.(1) This is somewhat surprising perhaps because aBMD cannot accurately measure the actual determinants of bone strength, such as the volumetric density or distribution of bone tissue.(2) Moreover, it is well known that aBMD is confounded by bone size and that the scans used clinically can also be distorted by aortic calcification and other artifacts in the older individuals at greatest risk of fracture. Previous work has shown that more refined measurements can be made by QCT,(3) which can quantify volumetric BMD (vBMD) of cortical, trabecular, and whole bone, as well as various aspects of bone structure and strength.(4) Bone microstructure can also be assessed, although in vivo measurements with the highest resolution are currently made by pQCT at the ultradistal radius or tibia.(5–8) In addition, we recently found that age- and sex-specific ratios for vertebral loading, relative to bone strength indices at the lumbar spine, resembled the incidence pattern of vertebral fractures over life more closely than did changes in spine vBMD.(9) Finally, it is now possible to estimate bone strength directly with finite element (FE) analysis at the lumbar spine.(10) This allows calculation of the factor-of-risk (ϕ, the ratio of applied bone loads to the bone failure load)(11) at the lumbar spine in vivo, which might improve fracture prediction over the use of bone strength parameters alone. The purpose of this report was to use virtually all of the available techniques (bone density, bone geometry, bone microstructure, bone strength, and spine load to bone strength ratios) in a population sample of postmenopausal women to explore the basis for vertebral fracture risk.

MATERIALS AND METHODS

Study subjects

After approval by Mayo Clinic's Institutional Review Board, subjects were recruited from age-stratified random samples of Rochester, MN, men and women selected from the medical records linkage system of the Rochester Epidemiology Project.(12) This analysis focused on the 248 postmenopausal women (mean age, 67.5 ± 12.0 yr; range, 39–97 yr) sampled in 2000 ± 1 yr.(4) Ninety-eight percent of them were white, reflecting the ethnic composition of the population (90% white in 2000). From this group, we identified all women who had experienced a clinically diagnosed fracture of the thoracolumbar spine caused by minimal or moderate trauma based on review of their inpatient and outpatient medical records in the community. Forty women met stringent criteria for vertebral fracture (see below), and they were frequency matched to an equal number of study women with no history of a prior hip, spine, or wrist fracture; too few potential controls met this criterion to allow individual age-matching to the cases. In addition, we studied 20 young normal women 21–29 yr of age from the larger study population. All subjects provided written informed consent before participation in the study.

Fracture ascertainment

Thoracic and lumbar spine fractures were assessed from the lateral QCT scout films by the study radiologist (PAR) according to the semiquantitative method.(13) Vertebrae with a grade 1 deformity by this scheme were also subjected to morphometry.(14) The confirmed vertebral fracture cases included 4 women with one or more grade 3 deformities, 13 with at least one grade 2 deformity, 20 with multiple grade 1 deformities, and 3 women with an isolated grade 1 deformity by semiquantitative criteria that also met 20% morphometric criteria for fracture (i.e., a ratio of anterior to posterior vertebral heights < 0.80). Altogether, 14 women had fractures that involved both the thoracic and lumbar spine, whereas 20 had thoracic deformities alone and 4 had only lumbar spine deformities. None of the control subjects had a prior clinical diagnosis of vertebral fracture, and none had even one grade 1 deformity that met the 20% morphometric criterion.

Bone density and structure measurements

As previously described,(4) vBMD and structure were assessed by single-energy QCT. Two different devices were used over the course of the study, a four-channel multidetector-row scanner (LightSpeed QX/i) and a comparable eight-channel system (LightSpeed Ultra), both from General Electric Medical Systems (Waukesha, WI, USA). Because cross-over studies were conducted on a phantom (European Spine Phantom; QRM, Möhrendorf, Germany), rather than subjects, we adjusted for the device used in some analyses. In addition to total vBMD of the L1–L3 vertebral bodies, we also measured trabecular vBMD in the central 70% of the midportion of the same vertebral body that was assessed by FE modeling (see below). Similar measurements were made in the proximal femur. A number of bone macrostructure measurements were made at midvertebral height, including total cross-sectional area, moment-of-inertia, section modulus, and “apparent” cortical thickness, recognizing that true cortical thickness is overestimated in the lumbar spine because of volume-averaging artifacts.(15)

We had no microstructure data for the spine but, instead, evaluated the nondominant wrist using high-resolution pQCT with a prototype of the Xtreme CT (Scanco Medical AG, Bassersdorf, Switzerland) on a subset of subjects at a later date. As described elsewhere,(6) bone tissue volume to total volume (BV/TV) was first derived from trabecular vBMD. Trabecular number (Tb.N) was taken as the inverse of the mean spacing of the 3D ridges (center points of the trabeculae) as detected in gray-level images.(16) Trabecular thickness was derived as BV/TV ÷ Tb.N, and trabecular separation (Tb.Sp) as 1 − BV/TV ÷ Tb.N, as in standard histomorphometry.(17) Tb.Sp.SD, the SD of Tb.Sp, is a measure of intraspecimen variation.

Finally, we measured aBMD of the spine region on a DXA whole body scan using the Prodigy instrument (GE Medical Systems, Madison, WI, USA) and software version 6.10.029. We previously showed that such scans are equivalent to dedicated lumbar spine DXA measurements in women, with r2 = 0.84 and an error in predicting lumbar spine aBMD of 6.5%.(18) In addition, aBMD of the total hip was estimated from the QCT hip scans using commercial software (QCT PR0; Mindways Software, Austin, TX, USA).

Estimation of vertebral strength characteristics

As described in detail elsewhere, we used the QCT data to calculate axial rigidity (AE), an index of resistance to compressive forces,(8) and flexural rigidity (EI), an index of resistance to forward bending moments,(19) in the lumbar spine. To estimate vertebral body failure loads (∼ strength) more directly, the QCT data were used to create voxel-based FE analyses in each subject (O.N. Diagnostics, Berkeley, CA, USA). Both the QCT and FE estimates were based on the same unfractured vertebral body: L3 (35 cases and 39 controls) or, if this were unsuitable for analysis, L2 (4 cases and 1 control) or, if necessary, L1 (l case). Briefly, each vertebral image (less posterior elements) was rotated into a standard coordinate system, thresholded, and converted into a 1 × 1 × 1.5-mm3 voxel-type FE mesh (Fig. 1), using eight-noded brick elements as described elsewhere.(20) Using a virtual layer of polymethylmethacrylate at the ends of each vertebral model to simulate compression strength testing of cadaver vertebrae,(21) uniform compressive displacement boundary conditions were applied. The compressive strength (N) of the vertebra was computed as the total reaction force generated at an imposed displacement equivalent to an overall bone compressive strain of 2%; this technique provides excellent measures of whole bone strength.(22)

Figure FIG. 1..

Voxel-based 3D finite element models of the L3 vertebrae (posterior elements removed) for a typical case (top) and control (bottom). The respective values of whole vertebral strength, trabecular strength, and integral vBMD were 4656 vs. 6095 N, 1340 vs. 3025 N, and 162 vs. 202 mg/cm3, respectively. Cutaway views show axial cross-sectional slices 3 mm below the superior endplate and at midsection. Colors represent the heterogeneity of the material strength applied to each voxel. Cross-sections are correctly scaled with respect to each other but are smaller than the 3D representations. Element sizes are 1 mm each side in-plane and 1.5 mm high. The white layers represent thin layers of polymethylmethacrylate virtually placed over the endplates for uniform load application. Models were loaded in the vertical direction.

As described previously,(20) each model was varied in a controlled fashion, and the compressive strength simulation was rerun on the altered models to compute other outcome variables. For example, intravertebral bone density variations were removed by applying the vertebra-specific vBMD uniformly across all voxels of the FE mesh and computing the resulting “homogenized density” strength. Trabecular strength was estimated by removing the outer 2 mm of bone from the model and recomputing strength for the remaining trabecular compartment. Because of boundary effects, adjacent trabecular bone is also unloaded, but the overall reduction in stiffness is comparable to that observed directly when the thin cortical shell (<0.5 mm) is removed in μCT FE models.(23) The strength associated with the cortical bone was not derived directly; instead, the difference between the whole vertebra and trabecular strengths was taken to represent the strength associated with the outer 2 mm of bone. Finally, to provide an assessment of the response to anteroposterior (AP) bending loads, a pure bending rotation of 1° was applied to the top endplate using linearly elastic analysis.(24) Strength in response to bending was calculated by combining the compressive strength and the ratio of axial stiffness to bending stiffness, along with the AP dimension of the vertebra at midheight, assuming a simple beam theory model.

Estimation of spinal loading

Only the axial compressive force was considered as the load applied to the vertebral body in this study. Because activities that lead to vertebral fracture are ill defined,(25) we chose to study several activities of daily living with varying compressive force magnitudes on L3: (1) upright standing (0° flexion); (2) forward flexion with the trunk at 45°; (3) forward flexion at 90°; and (4) forward flexion with the trunk at 90° while lifting 10 kg.(9) For upright standing, we assumed that the upper body weight is directly over the center of the vertebral column; therefore, there are no bending moments, and spinal muscle forces are absent. In the lifting activity, we assumed the arms hung straight down from the shoulders and that a 10-kg weight was held equally in the right and left hands (i.e., in the center of the body).

Factor-of-risk for vertebral fracture (ϕ)

The factor-of-risk is a biomechanical concept that relates the strength of a structure relative to the loads it is expected to carry. Theoretically, when ϕ ≥ 1, a fracture is predicted to occur, whereas no fracture is predicted when ϕ < 1.(26) In this study, ϕ was computed as the ratio of spinal compressive force to vertebral failure load for each activity. Because of possible errors in estimating both the spinal compressive forces and the vertebral failure load, the absolute fracture threshold is difficult to define. Nonetheless, higher values of ϕ indicate an increased risk of fracture.

Statistical analysis

Analyses were performed using SAS (SAS Institute, Cary, NC, USA) and Splus (Insightful Corp., Seattle, WA, USA). Bone variables were summarized using means and SD. T-scores were calculated using the means and SDs of the 20 young normal women. Pearson partial correlation coefficients were used to evaluate relationships between key bone density, structure, and strength variables, after adjusting for age.

The age-adjusted relative risk of fracture was estimated by ORs obtained from logistic regression models, where case status was the dependent variable. Potential predictors were assessed from the five main variable categories: bone density, bone geometry, bone microstructure, bone strength (all per SD decrease), and the factor-of-risk (per SD increase). Receiver operating characteristic (ROC) curves were also constructed for each variable from the logistic regression models.

RESULTS

The 40 postmenopausal women with a clinical history of vertebral fracture that was confirmed radiographically (mean ± SD age at assessment, 78.6 ± 9.0 yr) were frequency matched to 40 comparably aged women without a prior distal forearm, proximal femur, or vertebral fracture (age, 70.9 ± 6.8 yr). The most recent vertebral fracture had occurred 0–12 yr earlier (median age at fracture, 73 yr) as a result of no more than moderate trauma; none were caused by a specific pathological process (e.g., metastatic malignancy). Thirty-two percent of the fracture cases were currently on an osteoporosis-relevant drug, albeit not necessarily for osteoporosis treatment (7 on a bisphosphonate and 6 on estrogen) compared with 45% of controls (4 on a bisphosphonate, 11 on estrogen, and 3 on a selective estrogen receptor modulator [SERM]).

Because the fracture case and control women had similar mean heights (156 ± 7 versus 159 ± 5 cm; p = 0.849) and weights (67.8 ± 14.9 versus 70.3 ± 13.6 kg; p = 0.370), the estimated loads on their third lumbar vertebra were also similar at upright standing (392 ± 84 versus 395 ± 75 N; p = 0.182), at 45° forward flexion (1984 ± 464 versus 2013 ± 387 N; p = 0.186), at 90° forward flexion (2639 ± 635 versus 2706 ± 516 N; p = 0.173), and at 90° forward flexion while lifting 10 kg from the floor (3289 ± 636 versus 3337 ± 521 N; p = 0.191). Therefore, it was the skeletal variables that discriminated between fracture cases and controls. Note, however, that most T-scores in the five skeletal variable categories were greatly reduced in both groups compared with young normal women (Table 1).

Table Table 1.. Mean ± SD Values in the Five Main Variable Categories Comparing 40 Rochester, MN, Women With a Clinical Vertebral Fracture (Cases) to 40 Similarly Aged Women With No Osteoporotic Fracture (Controls)
original image

With regard to bone density variables, the vertebral fracture cases had 20% lower total lumbar spine vBMD (p = 0.018), after adjusting for age, compared with control women (Table 1). The discrepancy was somewhat greater for trabecular vBMD, which was 1.2 SD lower. In contrast, spine aBMD from the total body scan did not differ between cases and controls (p = 0.165). Total hip aBMD, estimated from hip QCT, was 10% less in the women with fractures.

Few geometric variables differed between cases and controls (Table 1). Thus, the women with vertebral fractures had slightly larger cross-sectional and endocortical areas but a significantly smaller apparent cortical thickness (1.6 ± 0.2 versus 1.8 ± 0.2 mm; p = 0.026). No significant differences were seen among the large number of microstructure variables assessed (some data not shown in Table 1) in the distal forearms of a subset of subjects (36 case and 34 control women).

With respect to bone strength variables, the women with vertebral fracture had significantly lower values for QCT-estimated axial rigidity, which was reduced by 19% among the cases compared with the 23% reduction seen in overall FE vertebral compressive strength (Table 1). The failure load of the “cortical” region (outer 2 mm layer of bone) under compression was 16% lower among the fracture cases compared with the reduction in trabecular compressive strength of 31%. The latter, in turn, related to deficits in vBMD, because the bone strength per unit vBMD was only 11% lower among the women with a vertebral fracture (p = 0.056). In addition to axial loads, the cases also had less strength under an AP bending moment as evaluated in the FE analysis. However, the ratio of bending to compressive strengths did not differ from the controls.

Because the applied loads on L3 and overall vertebral body strength were both expressed in Newtons, it was possible to obtain an estimate of the factor-of-risk under the different spinal loading conditions. Fracture cases had significantly greater (worse) values of ϕ in every instance: ϕ was worse among the cases whether the spinal load was upright standing, although fracture risk in that condition was judged to be very low, or bending 90° while lifting 10 kg (Table 1).

The association of different bone density, structure, and strength variables with vertebral fracture risk is delineated in Table 2. In age-adjusted logistic regression analyses within each of the five main variable categories, the strongest predictors of fracture risk were as follows: bone density (total vertebral vBMD: OR, 2.2; 95% CI, 1.1–4.3); bone geometry (vertebral apparent cortical thickness: OR, 2.1; 95% CI, 1.1–4.1); bone microstructure (no variables were significant, but these measurements were all made in the distal radius); and bone strength (“cortical” [outer 2 mm] compressive strength: OR, 2.5; 95% CI, 1.3–4.8). Of the various estimates of ϕ, that under a 90° forward bending load relative to total vertebral compressive strength was associated with the greatest increase in fracture risk (OR, 3.2; 95% CI, 1.4–7.5). Adjusting for the QCT device used did not alter these results.

Table Table 2.. Relative Risk of Clinical Vertebral Fracture (Age-Adjusted ORs Within the Five Main Variable Categories) Among Rochester, MN, Women
original image

It was not possible from this preliminary analysis to identify the “best” predictor of fracture risk because the different parameters were strongly correlated (Table 3). Thus, the age-adjusted correlation between lumbar spine vBMD and overall compressive vertebral strength by FE analysis was 0.86 (p < 0.001), and it would take 100 each of cases and controls to show that one of these results was 6% better than the other (α, 0.05; β, 0.80). The correlation is reflected in the fact that the areas under the ROC curves ranged only from 0.73 to 0.82 for the host of variables listed in Table 2. In addition, the selected variables delineated in Table 3 all correlated significantly with spine aBMD from the total body scan, except for ϕ (p = 0.066). However, each of them was a stronger predictor of vertebral fracture risk than spine aBMD (or total hip aBMD) in logistic regression analyses.

Table Table 3.. Age-Adjusted Correlations (Pearson Partial r) Between Key Parameters in Each of the Five Main Variable Categories, Along With Spine aBMD, Among 40 Rochester, MN, Women With a Clinically Diagnosed Vertebral Fracture
original image

DISCUSSION

Despite its limitations, aBMD by DXA is widely used in clinical practice for the diagnosis and management of osteoporosis. This preliminary analysis identified other strong predictors of vertebral fracture risk, and this additional information sheds light on the association of aBMD with vertebral fractures. Thus, spine aBMD was strongly correlated with total lumbar spine vBMD, which in turn was highly correlated with overall vertebral compressive strength by FE (age-adjusted partial r = 0.86 in cases and 0.91 in controls; both p < 0.001). In excised bones, the correlation with whole vertebra failure load is greater for FE-modeled bone strength (r2 = 0.86) than for vertebral trabecular vBMD by QCT (r2 = 0.53).(10) In this study, even the correlation of spine aBMD with overall vertebral compressive strength was 0.61 (p < 0.001), and ex vivo analyses have also found high correlations between vertebral failure load and aBMD in women.(27–30) Note, however, that aBMD correlated least well with ϕ, suggesting that a comparison of bone loads to bone strength provides complementary information.

Of particular interest is the greater relative deficit in trabecular vBMD in the fracture cases that may reflect disordered trabecular bone microstructure. Boutroy et al.(5) found that a 26% increase in Tb.Sp.SD, a measure of variation in trabecular spacing in the distal forearm, provided the best discrimination between mixed fracture patients and controls, and others have found similar results.(7) We found an even greater 36% increase in forearm Tb.Sp.SD among the women with a vertebral fracture, but this did not prove to be an independent predictor of fracture risk because of our limited sample size. Although microstructure at one skeletal site may not correlate closely with microstructure at another site,(31) variability of bone density within the vertebral body does seem to be an important risk factor.(20,32) Others have shown that trabecular disruption, trabecular variation, and loss of trabecular number are likely to have a particularly adverse influence on bone strength in the vertebrae.(32–34) Indeed, when we corrected intravertebral variability in bone density in the FE analysis (i.e., by assigning the average density in the vertebral body to every voxel), overall compressive strength was estimated to increase by 21% (from 5746 to 6958 N) in controls and by 31% (from 4410 to 5797 N) in the fracture cases.

However, cortical bone also seems to be important. By estimating overall vertebral strength before and after virtually removing the peripheral 2-mm layer of bone in the FE models, it was possible to show that the cortical compartment carried about one half of the compressive load in both cases and controls, as found also in more detailed analyses of cadaver vertebrae.(35) Even though the strength of this cortical compartment was reduced by only 16% among the women with vertebral fracture compared with a 31% reduction in trabecular compressive strength, the relative fracture risks associated with a 1 SD decrease in either parameter were similar.

Spine aBMD is confounded by bone size,(36) which is an independent determinant of vertebral strength.(24,37–39) We previously showed an age-related 14% increase in vertebral cross-sectional area among women in this population,(4) which should be protective in light of the fact that vertebral compressive forces declined slightly over life in conjunction with age-related weight loss.(9) Nevertheless, as judged relative to a derived estimate of bone strength, a marked age-related increase in ϕ (and predicted fracture risk) at the lumbar spine was seen; this was attributed to an age-related decline in vertebral vBMD, which was more extensive in women than in men (−55% versus −39%).(9) This study reveals that modestly greater cross-sectional area in the vertebral fracture cases was accompanied by greater endocortical area and lower apparent cortical thickness, as well as reduced compressive strength in the outer 2 mm of bone.

This study further confirmed the value of combining data on applied loads with failure loads. Thus, the ratio of the compressive load applied to L3 with a 90% forward bend compared with overall vertebral compressive strength was 36% worse on average among the women with a vertebral fracture. This represented a greater discrepancy than either the load (<3% lower) or the strength variables (23% lower) alone. Moreover, the ability to discriminate fractures improved to 3.2 per SD increase in ϕ relative to the odds ratio of 2.2 per SD decrease observed for overall vertebral compressive strength. These results are much more encouraging than those found with the analogous fracture risk index (FRI), where different methods (e.g., spine DXA) were used to estimate applied and failure loads.(40)

Our study had a number of strengths. First, the subjects were recruited from an age-stratified random sample of community women. Although the cross-sectional assessment of vertebral fractures is problematic,(41) the cases had all been clinically diagnosed with a vertebral fracture, and these were subsequently confirmed by the widely accepted semiquantitative technique.(13) The control women had not even a single grade 1 anterior wedge deformity that met the 20% morphometric criteria; lesser deformities probably represent statistical artifacts.(41) Both cases and their community controls were evaluated with state-of-the-art central QCT in the spine and high-resolution pQCT in the distal radius. Also, we estimated vertebral failure load (∼ strength) both from QCT-derived parameters and from more detailed FE analyses. The latter approach is restricted to the evaluation of isolated vertebral bodies, but it has provided excellent predictions of vertebral compressive strength as measured directly in cadaver studies.(10) Finally, we related bone strength to a range of spine loads encountered routinely in daily life, although our focus was on simplified compressive loads.

There are also some limitations. In particular, we had available only a limited number of postmenopausal women with a vertebral fracture. This precluded us from identifying the “best” determinant of vertebral fracture risk among those studied, nor could we evaluate the influence on bone structure of more estrogen and SERM use among controls than cases (14 versus 6). A larger group of vertebral fracture cases and controls is now being recruited from the general population, but results from that study will not be available for some time. We also lack information about the specific activities that actually led to these fractures. Instead, we estimated vertebral compressive forces during bending and lifting using a simple model of the trunk.(9) In addition, the comparison with DXA is only provisional because we estimated total hip aBMD from the hip QCT scan and lumbar spine aBMD from the spine region of a total body scan, although we have shown that the latter is highly correlated with dedicated lumbar spine DXA values in women.(18) Finally, we only assessed bone microstructure at a distant site, the ultradistal radius, but it may be possible in the future to evaluate microstructure in the lumbar spine.(42)

As noted in the Introduction, aBMD by DXA predicts fracture risk and is correlated with vertebral compressive strength. However, DXA does not directly measure any of the components (volumetric bone density, bone geometry, bone microstructure, bone strength) that were linked most strongly to vertebral fracture risk in this analysis. On the other hand, most of these same bone density, structure, and strength variables were highly correlated with spine aBMD. A much larger study will be required to quantify the actual improvement in fracture risk prediction that might follow the substitution of central QCT, with its ability to measure additional risk parameters, for lumbar spine DXA. More important, perhaps, is the addition of bone loading estimates to the skeletal strength parameters, because combinations of the two were more strongly associated with vertebral fracture risk than densitometry measures alone. By allowing an analysis of bone structure and density interactions under different loading conditions, the greater specificity of these new variables should provide better insights into the pathogenesis of vertebral fractures.

Acknowledgements

This work was supported by Research Grants R01-AR27065 and M01-00585 from the National Institutes of Health, U.S. Public Health Service. The authors thank Margaret Holets for the peripheral QCT measurements, Lisa McDaniel, RN, and Louise McCready, RN, for assistance in recruitment and management of the study subjects, James Peterson for assistance with data management and file storage, and Mary Roberts for assistance in preparing the manuscript.

Ancillary